What Device and TCAD Engineers Do

Every other path on this site works at one remove from the physical transistor: RTL design assumes a cell library already exists; physical design places and routes cells whose electrical behavior is already captured in a Liberty file; STA signs off timing using those same already-characterized cells. This path is about the layer underneath all of that: the people and tools that decide what the transistor itself is, predict how it will behave before it is ever built, and turn that prediction into the compact model every other tool in the flow ultimately trusts.

This lesson is a map of the role and the tools, so the physics lessons that follow (2-5) and the simulation lessons after them (6-9) have somewhere to attach.

Reading time: about 7 minutes.

1.1 Two connected disciplines

"Device and TCAD engineering" covers two closely linked jobs, often done by the same team or even the same engineer across a device's development:

• Device engineering asks: given a target technology node, what transistor structure, dimensions and doping profile deliver the performance (drive current, leakage, variability) the node's roadmap requires? This work spans experimental silicon characterization and simulation together.
• TCAD (technology computer-aided design) is the simulation side specifically: software that predicts, before a wafer is ever processed, what a proposed fabrication sequence will produce (process simulation, lesson 6) and how the resulting structure will electrically behave (device simulation, lesson 7). TCAD lets a team explore many candidate structures and process options computationally, far faster and cheaper than running silicon for each one, narrowing the field before anything is actually fabricated.

Commercial TCAD suites (Synopsys Sentaurus and Silvaco Victory are two well-known examples, named here only as examples of what this category of tool looks like, not as an endorsement of either) provide both process and device simulation in an integrated flow, so a simulated process sequence's output structure feeds directly into the device simulator without the engineer having to redraw it by hand.

1.2 Where this sits in the bigger picture

A device and TCAD team's output feeds forward into two different audiences:

• Process integration: translating a device's target structure into an actual, manufacturable sequence of the steps covered across this whole Manufacturing section (lithography, etch, deposition, implant, CMP), coordinating with the equipment and process-control teams covered in the Fab Equipment & Yield path.
• Design enablement: once a device is characterized, compact models (lesson 9) extracted from it are packaged into the process design kit (PDK) that RTL, synthesis and physical design teams use every day, without ever needing to think about the device physics underneath a standard cell's Liberty timing arcs.

These two audiences are not served sequentially; in practice they run in parallel for much of a node's development, because a device structure's manufacturability and its ultimate design-enablement value are both evaluated against the same candidate options, and a structure that looks promising electrically but turns out hard to integrate into a reliable, repeatable process flow is exactly the kind of finding that needs to come back early, not after a PDK has already been published to design teams.

1.3 A typical technology development cycle

A new technology node does not go from a blank page directly to a finished PDK. A representative cycle looks roughly like this: device engineers and architects first set target specifications for the node (drive current, leakage, density targets relative to the previous node); TCAD engineers propose candidate device structures and process sequences and evaluate them computationally, using process simulation (lesson 6) to predict the structure each candidate sequence would produce and device simulation (lesson 7) to predict how that structure would behave electrically, narrowing many candidates down to a short list worth actually building. Early silicon lots are then run for the most promising candidates, process integration (coordinating the real fabrication sequence across the equipment covered in the Fab Equipment & Yield path) turns the TCAD-evaluated sequence into an actual qualified process flow, and the resulting real devices are measured and compared back against what TCAD predicted. Discrepancies between prediction and measurement feed back into refining the TCAD models themselves (lesson 7.1's drift-diffusion model, and the process models in lesson 6, both have parameters that get calibrated against real measured data, not just assumed from first principles alone), tightening the loop for the next iteration.

Only once a device structure and process flow are both qualified does parameter extraction (lesson 9) fit the final compact model that ships in the PDK. This loop, propose, simulate, build, measure, calibrate, repeat, commonly runs for many months to a few years before a node's PDK is considered mature enough for broad design use, which is also why an early PDK revision for a brand-new node is typically treated as less stable and more likely to change than a mature node's PDK that many previous designs have already been taped out against.

1.4 Why this path matters even if you never touch TCAD

A physical design, STA or low-power engineer will likely never open a TCAD tool directly. This path is still useful because several things taken for granted in the rest of the flow come directly from here: why threshold voltage, subthreshold swing and short-channel effects (lesson 4) appear in a Liberty timing arc at all; why variability and reliability margins (lesson 8) exist in signoff corners; why a cell library has multiple flavors at different drive strengths and threshold voltages, each one a different device design choice made upstream. Understanding where these numbers come from makes signoff margins and corner definitions easier to reason about, rather than treated as arbitrary constants handed down from the PDK.

1.5 Interview questions

What is the difference between device engineering and TCAD, in one sentence each? Device engineering decides what transistor structure and process will meet a node's performance targets; TCAD is the simulation software and methodology used to predict process and device behavior before fabricating anything.

What two downstream audiences does device and TCAD work feed? Process integration (turning the target structure into an actual manufacturable process sequence) and design enablement (packaging the characterized device into PDK compact models for RTL/synthesis/PD).

Why does TCAD simulate before fabricating real silicon? Simulating many candidate structures and process options computationally is far faster and cheaper than running silicon for each one, letting the team narrow the field before committing to actual fabrication.

Name the two simulation types a TCAD suite integrates. Process simulation (predicts the structure a process sequence produces) and device simulation (predicts the electrical behavior of that structure).

Why should a physical design or STA engineer care about device and TCAD work, even without using the tools directly? Timing arcs, threshold voltages, signoff corners and cell library variants all trace back to device-level choices made in this discipline; understanding the origin makes those numbers easier to reason about rather than treated as opaque PDK constants.